diff --git a/tests/test_client.py b/tests/test_client.py
index ebbfa2e..204c256 100644
--- a/tests/test_client.py
+++ b/tests/test_client.py
@@ -3,7 +3,13 @@
 import arxiv
 from datetime import datetime, timedelta
 from pytest import approx
+from requests import Response
 
+def empty_response(code: int) -> Response:
+    r = Response()
+    r.status_code = code
+    r._content = b''
+    return r
 
 class TestClient(unittest.TestCase):
     def test_invalid_format_id(self):
@@ -90,10 +96,10 @@ def test_no_duplicates(self):
             self.assertFalse(r.entry_id in ids)
             ids.add(r.entry_id)
 
+    @patch('requests.Session.get', return_value=empty_response(500))
     @patch("time.sleep", return_value=None)
-    def test_retry(self, patched_time_sleep):
-        broken_client = TestClient.get_code_client(500)
-
+    def test_retry(self, mock_sleep, mock_get):
+        broken_client = arxiv.Client()
         def broken_get():
             search = arxiv.Search(query="quantum")
             return next(broken_client.results(search))
@@ -109,77 +115,73 @@ def broken_get():
                 self.assertEqual(e.status, 500)
                 self.assertEqual(e.retry, broken_client.num_retries)
 
+    @patch('requests.Session.get', return_value=empty_response(200))
     @patch("time.sleep", return_value=None)
-    def test_sleep_standard(self, patched_time_sleep):
-        client = TestClient.get_code_client(200)
+    def test_sleep_standard(self, mock_sleep, mock_get):
+        client = arxiv.Client()
         url = client._format_url(arxiv.Search(query="quantum"), 0, 1)
         # A client should sleep until delay_seconds have passed.
         client._parse_feed(url)
-        patched_time_sleep.assert_not_called()
+        mock_sleep.assert_not_called()
         # Overwrite _last_request_dt to minimize flakiness: different
         # environments will have different page fetch times.
         client._last_request_dt = datetime.now()
         client._parse_feed(url)
-        patched_time_sleep.assert_called_once_with(approx(client.delay_seconds, rel=1e-3))
+        mock_sleep.assert_called_once_with(approx(client.delay_seconds, rel=1e-3))
 
+    @patch('requests.Session.get', return_value=empty_response(200))
     @patch("time.sleep", return_value=None)
-    def test_sleep_multiple_requests(self, patched_time_sleep):
-        client = TestClient.get_code_client(200)
+    def test_sleep_multiple_requests(self, mock_sleep, mock_get):
+        client = arxiv.Client()
         url1 = client._format_url(arxiv.Search(query="quantum"), 0, 1)
         url2 = client._format_url(arxiv.Search(query="testing"), 0, 1)
         # Rate limiting is URL-independent; expect same behavior as in
         # `test_sleep_standard`.
         client._parse_feed(url1)
-        patched_time_sleep.assert_not_called()
+        mock_sleep.assert_not_called()
         client._last_request_dt = datetime.now()
         client._parse_feed(url2)
-        patched_time_sleep.assert_called_once_with(approx(client.delay_seconds, rel=1e-3))
+        mock_sleep.assert_called_once_with(approx(client.delay_seconds, rel=1e-3))
 
+    @patch('requests.Session.get', return_value=empty_response(200))
     @patch("time.sleep", return_value=None)
-    def test_sleep_elapsed(self, patched_time_sleep):
-        client = TestClient.get_code_client(200)
+    def test_sleep_elapsed(self, mock_sleep, mock_get):
+        client = arxiv.Client()
         url = client._format_url(arxiv.Search(query="quantum"), 0, 1)
         # If _last_request_dt is less than delay_seconds ago, sleep.
         client._last_request_dt = datetime.now() - timedelta(seconds=client.delay_seconds - 1)
         client._parse_feed(url)
-        patched_time_sleep.assert_called_once()
-        patched_time_sleep.reset_mock()
+        mock_sleep.assert_called_once()
+        mock_sleep.reset_mock()
         # If _last_request_dt is at least delay_seconds ago, don't sleep.
         client._last_request_dt = datetime.now() - timedelta(seconds=client.delay_seconds)
         client._parse_feed(url)
-        patched_time_sleep.assert_not_called()
+        mock_sleep.assert_not_called()
 
+    @patch('requests.Session.get', return_value=empty_response(200))
     @patch("time.sleep", return_value=None)
-    def test_sleep_zero_delay(self, patched_time_sleep):
-        client = TestClient.get_code_client(code=200, delay_seconds=0)
+    def test_sleep_zero_delay(self, mock_sleep, mock_get):
+        client = arxiv.Client(delay_seconds=0)
         url = client._format_url(arxiv.Search(query="quantum"), 0, 1)
         client._parse_feed(url)
         client._parse_feed(url)
-        patched_time_sleep.assert_not_called()
+        mock_sleep.assert_not_called()
 
+    @patch('requests.Session.get', return_value=empty_response(500))
     @patch("time.sleep", return_value=None)
-    def test_sleep_between_errors(self, patched_time_sleep):
-        client = TestClient.get_code_client(500)
+    def test_sleep_between_errors(self, mock_sleep, mock_get):
+        client = arxiv.Client()
         url = client._format_url(arxiv.Search(query="quantum"), 0, 1)
         try:
             client._parse_feed(url)
         except arxiv.HTTPError:
             pass
         # Should sleep between retries.
-        patched_time_sleep.assert_called()
-        self.assertEqual(patched_time_sleep.call_count, client.num_retries)
-        patched_time_sleep.assert_has_calls(
+        mock_sleep.assert_called()
+        self.assertEqual(mock_sleep.call_count, client.num_retries)
+        mock_sleep.assert_has_calls(
             [
                 call(approx(client.delay_seconds, abs=1e-2)),
             ]
             * client.num_retries
         )
-
-    def get_code_client(code: int, delay_seconds=0.1, num_retries=3) -> arxiv.Client:
-        """
-        get_code_client returns an arxiv.Cient with HTTP requests routed to
-        httpstat.us.
-        """
-        client = arxiv.Client(delay_seconds=delay_seconds, num_retries=num_retries)
-        client.query_url_format = "https://teapot.fly.dev/{}?".format(code) + "{}"
-        return client